Publications

Recent advances in neural recording techniques open exciting possibilities of better understanding whole populations of neurons. Devices such as APS MEA (Active Pixel Sensor Microelectrode Array) [1, 2] allow for simultaneous recordings from 4096 channels (64x64 grid) at near-cellular resolution (electrode size: 21μm, electrode spacing: 42μm) and constitute a potentially very rich and detailed source of information on the dynamics of neural systems. Such volumes of data are however difficult to analyse: simple measures such as mean firing rates and correlations are often insufficient to capture interesting phenomena, while more sophisticated approaches can be computationally intensive and hard to interpret. Here we examine the applicability of pairwise maximum entropy (MaxEnt) [3–5] modelling to describe APS MEA data.

Our current understanding of the dynamics of neural circuits is limited by the poor resolution of multi-neuron recordings from large neural populations, which largely prevents the experimental verification of theoretical models and predictions. It is, for instance, difficult to distinguish between different potential classes of network architecture, such as feed-forward or recurrent networks, on the basis of simultaneous recordings from just tens of neurons. Recent advances in electronics have now made it possible to simultaneously record from thousands of neurons.

In this work, we investigate the spontaneous bursting behaviour expressed by in vitro hippocampal networks by using a high-resolution CMOS-based microelectrode array (MEA), featuring 4096 electrodes, inter-electrode spacing of 21 µm and temporal resolution of 130 µs. In particular, we report an original development of an adapted analysis method enabling us to investigate spatial and temporal patterns of activity and the interplay between successive network bursts (NBs). We first defined and detected NBs, and then, we analysed the spatial and temporal behaviour of these events with an algorithm based on the centre of activity trajectory. We further refined the analysis by using a technique derived from statistical mechanics, capable of distinguishing the two main phases of NBs, i.e. (i) a propagating and (ii) a reverberating phase, and by classifying the trajectory patterns. Finally, this methodology was applied to signal representations based on spike detection, i.e. the instantaneous firing rate, and directly based on voltage-coded raw data, i.e. activity movies. Results highlight the potentialities of this approach to investigate fundamental issues on spontaneous neuronal dynamics and suggest the hypothesis that neurons operate in a sort of 'team' to the perpetuation of the transmission of the same information.

Based on experiments performed with high-resolution Active Pixel Sensor microelectrode arrays (APS-MEAs) coupled with spontaneously active hippocampal cultures, this work investigates the spatial resolution effects of the neuroelectronic interface on the analysis of the recorded electrophysiological signals. The adopted methodology consists, first, in recording the spontaneous activity at the highest spatial resolution (interelectrode separation of 21 μm) from the whole array of 4096 microelectrodes. Then, the full resolution dataset is spatially downsampled in order to evaluate the effects on raster plot representation, array-wide spike rate (AWSR), mean firing rate (MFR) and mean bursting rate (MBR). Furthermore, the effects of the array-to-network relative position are evaluated by shifting a subset of equally spaced electrodes on the entire recorded area. Results highlight that MFR and MBR are particularly influenced by the spatial resolution provided by the neuroelectronic interface. On high-resolution large MEAs, such analysis better represent the time-based parameterization of the network dynamics. Finally, this work suggest interesting capabilities of high-resolution MEAs for spatial-based analysis in dense and low-dense neuronal preparation for investigating signaling at both local and global neuronal circuitries.

This paper presents a chip-based electrophysiological platform enabling the study of micro- and macro-circuitry in in-vitro neuronal preparations. The approach is based on a 64 × 64 microelectrode array device providing extracellular electrophysiological activity recordings with high spatial (21 µm of electrode separation) and temporal resolution (from 0.13 ms for 4096 microelectrodes down to 8 µs for 64 microelectrodes). Applied to in-vitro neuronal preparations, we show how this approach enables neuronal signals to be acquired for investigating neuronal activity from single cells and microcircuits to large scale neuronal networks. The main elements of the platform are the metallic microelectrode array (MEA) implemented in Complementary Metal Oxide Semiconductor (CMOS) technology similar to a light imager, the in-pixel integrated low-noise amplifiers (11 µV_rms) and the high-speed random addressing logic. The chip is combined with a real-time acquisition system providing the capability to record at 7.8 kHz/electrode the whole array and to process the acquired signals.

The spike represents the fundamental bit of information transmitted by the neurons within a network in order to communicate. Then, given the importance of the spike rate as well as the spike time for coding the activity generated at the level of a cell assembly, a relevant issue in extracellular electrophysiology is the correct identification of the spike in multisite recordings from brain areas or neuronal networks. In this paper, we present a novel spike detection algorithm, named Precise Timing Spike Detection (PTSD), aimed at (i) reducing the number of false positives and false negatives, in order to optimize the rate code, and (ii) improving the time precision of the identified spike, in order to optimize the spike timing. The PTSD algorithm considers consecutive portions of the signal and looks for the Relative Maximum/Minimum whose peak-to-peak amplitude is above a defined differential threshold and responds to specific requirements. To validate the algorithm, the presented spike detection has been compared with other methods either commercially available or proposed in the literature by using two benchmarking procedures: (i) visual inspection by a group of experts of a portion of signal recorded from a rat cortical culture and (ii) detection of the spikes generated by a realistic neuronal network model. In both cases our algorithm produced the best performances in terms of efficiency and precision. The ROC curve analysis further proved that the best results are reached by the application of the PTSD.

The microelectrode array (MEA) technology is continuously progressing towards higher integration of an increasing number of electrodes. The ensuing data streams that can be of several hundreds or thousands of Megabits/s require the implementation of new signal processing and data handling methodologies to substitute the currently used off-line analysis methods. Here, we present one approach based on the hardware implementation of a wavelet-based solution for real-time processing of extracellular neuronal signals acquired on high-density MEAs. We demonstrate that simple mathematical operations on the discrete wavelet transform (DWT) coefficients can be used for efficient neuronal spike detection and sorting. As the DWT is particularly well suited for implementation on dedicated hardware, we elaborated a wavelet processor on a field programmable gate array (FPGA) in order to compute the wavelet coefficients on 256 channels in real-time. By providing sufficient hardware resources, this solution can be easily scaled up for processing more electrode channels.

A platform for high spatial and temporal resolution electrophysiological recordings of in vitro electrogenic cell cultures handling 4096 electrodes at a full frame rate of 8 kHz is presented and validated by means of cardiomyocyte cultures. Based on an active pixel sensor device implementing an array of metallic electrodes, the system provides acquisitions at spatial resolutions of 42 microm on an active area of 2.67 mm x 2.67 mm, and in the zooming mode, temporal resolutions down to 8 micros on 64 randomly selected electrodes. The low-noise performances of the integrated amplifier (11 microV (rms)) combined with a hardware implementation inspired by image/video processing concepts enable high-resolution acquisitions with real-time preprocessing capabilities adapted to the handling of the large amount of acquired data.